Configuration panel

Within the side panel the user can load data and configure quality control options.

Load proteinGroups.txt input file

To begin the analysis the user has to upload the MaxQuant file (e.g.proteinGroups.txt), as specified above. After full upload of the file, rows that were only found in the reverse database, belonging to potential contaminants or that have only been identified by site are filtered automatically.

Quality control and data cleansing

• The user selects either LFQ or iBAQ as intensity metric to be considered for succeeding differential abundance analysis. If available, we suggest to use LFQ intensities as Eatomics was optimized for these. Internally, the intensity widget uses the selectProteinData function. In the case of iBAQ values, Eatomics performs automatic normalization via limma’s normalizeVSN() function. proteinGroups entries (rows) with a “+” in the “Reverse”, “Potential contaminant”, “Only identified by site” columns are removed automatically.
• The exclude column widget allows the user to exclude samples, especially if any outliers are found while conducting initial quality analysis such as PCA. Selecting a sample here, results in the removal of that sample from the consecutive steps analysis steps.
• To avoid proteins with many missing values across the samples, the user selects the minimum number of samples for which a protein must have been detected in. Internally the filter widget uses the filterProteins function.
• Meaningful gene names: As genes names are easier to interpret than peptide identifiers, the gene names are displayed primarily. As gene names can be non-unique, the user can choose to let Eatomics
  • prepare unique IDs for duplicate gene names or (make.unique() R base function)
  • to sum up multiple abundance values for one gene name (checkForIsoforms custom function). In the latter case, the user is informed about intensity shares. Within the data download, uniprot accession numbers and protein names (or any other supplied by the MaxQuant output) are given.
• Missing value imputation can be performed using knn (k-nearest-neighbour) from the impute package [5], MinDet or QRLIC from the imputeLCMD package [3] or a custom implementation of Perseus’ sampling from a down-shifted Gaussian distribution (implemented by Matthias Ziehm) with default parameters of width = 0.3 and shift = 1.8.

Load the sample description/clinical data file

Select and load the clinical data input file (e.g clinicaldata.txt), as specified above.

Visualization panel

In the main panel (right) interactive visualizations are shown.

Principal component analysis

A common method of dimensionality reduction is principal component analysis (PCA). Inherently, PCA calculates axes of most variation (principal components) within the abundance data. A common assumption is that a plot along the axes of most variation will segregate all samples/patients into groups under investigation. The user can choose which principle components to visualize in the PCA and can choose to color the samples based on the uploaded sample/clinical characteristics.

Distribution overview

The distribution overview gives an impression on the sample-wise distribution of all measured intensities.

Protein coverage

Protein coverage describes the count of distinct protein groups per sample.

Sample to sample heatmap

The sample-to-sample heatmap describes the biological and technical variability of the samples. The user can choose to use Euclidean distance or Pearson correlation as a (dis-) similarity metric. Formed clusters should resemble the sample groups under investigation.

Cumulative Protein Intensities

Protein intensities are cumulated across all samples and plotted according to their relative abundance. Colouring marks the respective quantile of the proteins. Highly abundant proteins, i.e., proteins ranked in the first quartile are colored in red and labels are specified. The top 20 ranked proteins and their cumulated intensity are given in the table to the right.

Experimental design configuration

• The clinical grouping factor widget allows the user to select the clinical parameter of their choice from the uploaded dataset.

• In case the first grouping factor is categorical, the compare widget shows all the available subgroups inside the selected clinical group. The user then selects two subgroups to compare and find the differential expressed genes in the selected subgroups.

• In case of the first grouping factor being numeric, the select threshold widget appears and asks the user to specify a threshold, e.g., in the case of age, the user would split the cohort into a young and and old group, based on the given age at which to split. Automatically the two resulting groups are compared.

• Alternatively, the option to use continuous response may be checked. Eatomics then ignores the threshold and performs the analysis based on the continuous value. Please take special caution on how to interpret the results.

• By checking the impute missing values box, Eatomics uses the imputed data using the method specified in the previous tab. However, in the background Eatomics performs a filter on imputed data, which leads to exclusion of proteins, for which there are fewer than 50 % of measured values in both compared groups available. As a result proteins may be found to be upregulated in one group although the protein was just not detected in the other group. If the box is not checked, the limma function does perform it’s own missing value imputation.

• Include covariates activates the selection of parameters from the sample description file to be included into the experimental design model. As many covariates as reasonable, i.e., factors that may influence the abundance, but are not of primary interest for the current analysis, can be selected.

• A checkmark in stratification or filter expands a selection ver similar to the first part of the module and leads Eatomics to consider further specification of the design.

• Along the line of the first part, a second parameter can be selected. The second paramater must be categorical, which means that a threshold for continuous parameters has to be selected. As a result, new combined groups appear and can be selected to resemble the stratified contrast of interest. In the case of the first parameter being continuous the second parameter can be considered as a filter which only includes the groups selected in the second part in the experimental design.

• The analyze button starts the analysis and should be used whenever a new design has been configured.

Visualization of differentially abundant proteins

The result of differential abundance analysis is displayed in an interactive volcano plot, two tables of up- and downregulated proteins and box and scatter plots of actual protein abundance.

Volcano plot

The volcano plot shows the log2 fold change value on the x-axis and the negative log10 of the Benjamini-Hochberg Benjamini-Hochberg adjusted p-value on the y-axis. Significant results are shown in yellow. The threshold of log2 fold change and p-values considered significant can be set by the user directly within the threshold box. A hover over a dot in the volcanoe plot will display the respective gene name. A positive fold change can be interpreted as that protein being higher in abundance in the first selected group when compared to the second, in the case of a categorical response. When a continuous response is modeled, the fold change has to be interpreted as the slope, i.e., increase (positive log2 fold change) or decrease (negative log2 fold change), of the protein abundance with regard to a change of one unit of the response variable. For example, if age is analyzed, a log2 fold change of -0.2 would mean a decrease of about 1.15 (2^0.2) in LFQ intensity and thus protein abundance with every year of age.

Result tables and box/scatter plots

Significant results are listed in two tables below the volcano plot. They show the actual logFC, p-value and the adjusted p-value. A click on a protein entry in the data table produces a box (in the case of a categorical response variable) or scatter plot (in the case of a continuous response variable) showing the actual abundance values of selected proteins with regard to the tested comparisons.
The color of individual dots can be chosen to reflect a parameter from the sample description file.

The user selects a gene set from the list of MsigDB (H-) Hallmark, (C5.all.-) Gene Ontology (GO) all terms or subsets of GO molecular function (C5.mf.), GO biological process (C5.bp.), or GO cellular compartment (C5.cc.), (C2.cp.reactome-) Reactome, (C2.cp.biocarta.-) Biocarta or (C2.cp.kegg.-) KEGG to calculate the enrichment score. Alternatively, the user may provide a custom gene set file in .gmt format by pasting it into the texttt{Data/GeneSetDBs} folder. As in the original code, many parameters may be set by an expert user. However, for a quick setup the default options from the original publication are implemented. For convenience, the user may specify a prefix for the output files. Enrichment scores are stored as .gct files the Data/EnrichmentScores folder to be accessed from the next tab. An alert message will be pop up on completion of calculation.

As a result, the UI is almost identical to step 2. As a difference, instead of using the prepared preotein abundance data, the user has to choose the enrichment score file from those prepared on the ssGSEA tab. The selected file should be the plain enrichment score file, so neither the file containing the p-values suffix, nor fdr-suffix file are appropriate. After configuring the experimental design, results are visualized in the interactive volcano plot and gene lists and box plots are available for further exploration as well. Significant results are gene sets, e.g. GO terms or pathways, that are enriched in the first experimental group versus the second group. Fold changes give an impression on the effect size, which is an advantage over other methods, which mainly deliver p-values to assess importance.

On both differential analysis tabs (step 2 and 4) there are buttons for displaying more detailed information on the configured experimental design and for the download of report and data tables.